Demonstration of UAV-Based GPS Jammer Localization

During a Live Interference Exercise

Adrien Perkins, Louis Dressel, Sherman Lo, Tyler Reid, Kazuma Gunning and Per Enge

Stanford University

BIOGRAPHIES

Adrien Perkins is a Ph.D. candidate in the Global

Positioning System (GPS) Laboratory at Stanford

University working under the guidance of Professor Per

Enge in the Department of Aeronautics and Astronautics.

He received his Bachelor of Science in Mechanical

Aerospace Engineering from Rutgers University in 2013

and his Master of Science in Aeronautics and Astronautics

from Stanford University in 2015.

Louis Dressel is a Ph.D. candidate in the Aeronautics and

Astronautics Department at Stanford University. There, he

works on a joint project with both the Stanford Intelligent

Systems Lab and the GPS Laboratory implementing

control algorithms for a jammer-hunting UAV. He

received his Bachelor of Science in Aerospace

Engineering with a minor in Computer Science from

Georgia Tech in 2013 and his Master of Science in

Aeronautics and Astronautics from Stanford University in

2015.

Sherman Lo is a senior research engineer at the Stanford

University GPS Laboratory.

Tyler Reid is a Ph.D. candidate in the GPS Laboratory at

Stanford University.

Kazuma Gunning is a Ph.D. candidate in the GPS

Laboratory at Stanford University.

Per Enge is a Professor in the Department of Aeronautics

and Astronautics at Stanford University, where he is the

Vance and Arlene Coffman Professor in the School of

Engineering. He directs the GPS Laboratory and the

Stanford Center for Position, Navigation and Time

(SCPNT).

ABSTRACT

To assist in the mitigation of the effects posed by Global

Navigation Satellite System (GNSS) jammers, this paper

demonstrates the use of an unmanned aerial vehicle (UAV)

capable of autonomously localizing the source of Global

Positioning System (GPS) jamming in a live jamming

exercise hosted by the Department of Homeland Security

(DHS).

Developing an autonomous UAV for jammer localization

in a real-world environment needs to address three main

challenges: accurate measurements of the jamming signal,

rapid localization steps and reliable navigation in the

presence of interference. Our system, Jammer Acquisition

with GPS Exploration and Reconnaissance (JAGER), has

been developed to address those main challenges for rapid

localization and has been previously tested with localizing

Wi-Fi signals. This paper outlines the modifications to

JAGER required to be able to move from localizing Wi-Fi

sources to localizing GPS jammers. Modifications include

new sensing equipment for determining the bearing to the

jammer and additional navigation systems to fly while the

jammer is active.

The main goal for the testing was to demonstrate the

feasibility of JAGER to localize a GPS jammer at realistic

distances and the performance of several different

localization methods. The second goal was to explore

possible GPS-denied navigation solutions.

INTRODUCTION

GPS has become a critical element in many different

industries ranging from commercial aviation to

telecommunications and even the power grid. The

ubiquitous nature of GPS today has brought with it a

growth in commercial jammers that, while illegal, are used

for personal privacy and pose a threat to today’s industries,

especially in the wrong hands. In addition to efforts to

toughen and augment GPS to combat jammers, the ability

to quickly interdict and eliminate an interfering jammer is

important. To address the protection against jammers, this

paper discusses the development and testing of Jammer

Acquisition with GPS Exploration and Reconnaissance

(JAGER), an autonomous multirotor unmanned aerial

vehicle (UAV) capable of localizing commercially

available GPS jammers, in the presence of on-air Global

Navigation Satellite System (GNSS) jamming during the

Department of Homeland Security (DHS) First Responder

Electronic Jamming Exercise (FREJE) conducted at a

Department of Defense (DOD) test range in 2016.

Our prior work demonstrated the capability of the UAV to

localize a Wi-Fi proxy jammer over a short distance [1].

Several major changes had to be made to enable

localization of a GPS jammer in a more typical operational

area including new sensing subsystem, faster algorithms

and navigation systems capable of operating in a GPS-

denied environment. We will describe these modifications

as well as the equipment changes and testing needed to

ensure safe and reliable test operations.

To ensure rapid localization of the jammer, measurements

of the jamming signal need to be accurate and therefore

robust to jamming signal variations and radio frequency

interference (RFI) not directly coming from the jammer.

This may be RFI from the other equipment in the vicinity,

noise from the vehicle and signal reflections from the

jammer. For the UAV presented in this paper, the sensor

used and configuration onboard the vehicle reduced effects

from the signal variations and RFI, and algorithmic

methods are used to determine bearing through the noise.

The primary measurement for localization is the bearing to

the GPS jammer, determined from a directional antenna

and leveraging the ease of rotation of a multirotor. To

contend with RFI noise, a robust algorithm to extract

bearing has been developed and is demonstrated through

flight-testing.

To handle the larger search area for these tests, instead of

using the computationally expensive Partially Observable

Markov Decision Process (POMDP) methods used

previously [1], the rapid localization of the jammer is

performed through several one-step optimal (greedy)

solutions that minimize a specified information-theoretic

objective.

Navigation in a GPS-denied environment is essential for a

system searching for a GPS jammer. For testing during the

live jamming exercise, our UAV integrated several

navigation solutions including: GLONASS, Locata and

GPS itself. Our equipment setup and the nature of the

jamming allowed us to conduct the flight trials using

GLONASS for positioning and control of the UAV. In fact,

a majority of the flights were conducted using the

GLONASS receiver as it provided a low weight, easy to

integrate navigation package. The presence of a jammer

does not mean that GPS/GNSS should be ignored,

however. Our test demonstrated that it is viable to localize

the jammer while maintaining a standoff distance where

GPS is not degraded by the interference.

In addition to describing the methods for localization for

an autonomous UAV, this paper presents the modifications

of our UAV to ensure safe and reliable operations of the

UAV in a live interference exercise. In assembling the

system, RFI was a significant challenge for all antennas in

close proximity onboard the UAV and care was taken to

minimize the interference between all systems onboard the

UAV. Furthermore, all flight-testing was performed at

night at long range so reliable visual markers and

communication systems were required.

TEST SETUP / CONFIGURATION

During the DHS FREJE both first responder groups and

academia where invited to participate in an exercise with

live interference across many different frequencies that can

affect first responders. For our testing, we focused

exclusively on jamming in the GPS L1 band with a DHS

provided commercial off-the-shelf (COTS) jammer.

Figure 1: Minutes of in air time for each night of testing

The exercise was a weeklong event, with flight-testing

occurring during the nights of Tuesday through Friday. An

outline of the overall flight time for each night of our

testing is shown in Figure 1. The figure shows the overall

total flight time, the amount of time GPS was unavailable

due to jamming and the amount of time JAGER was flying

autonomously in jammer hunting mode. For our test

campaign, the first two nights were spent on final

integration, ensuring RFI mitigation was properly taken

care of, and safety testing, both on the ground and in the

air. On the final two nights, Thursday and Friday, we

almost exclusively tested the various localization

algorithms and navigation systems. Hence hunt mode

represents a high percentage of the total flight time.

For each test, the COTS jammer was placed at a known

location and the UAV was started at another known

position to execute its search.

Jammer

DHS provided COTS handheld GPS jammer similar to the

one depicted in Figure 2.

Figure 2: Example handheld COTS GPS jammer

0

20

40

60

80

Tues Wed Thurs Fri

In Air Time

flight jammed hunting

The online specifications for the jammer state a power of 2

Watts (W) in the L1, L2 & L3, L4 and L5 bands, resulting

in an effective 0.5 W of interference in the GPS L1 band.

Since the localization sensor onboard the UAV senses

jamming in the L1 band, this is effectively a 0.5 W jammer

for flight-testing. However, it is not clear that this is the

actual radiated power as the jammer had several settings

that could be enabled through dual in-line package (DIP)

switches. The online specifications also state an effective

range of 5-15 meters (m) whereas analysis such as those

from [2] would suggest a range of approximately 20

nautical miles without mitigation. Our test experience

indicated something a range closer to several hundred

meters. This is discussed in the Jammer Performance

section below.

JAGER

JAGER consists of a modified DJI S1000 airframe with a

Pixhawk flight control computer and flight control code

customized to the platform and the mission [3]. The

Pixhawk also contains low-cost microelectromechanical

systems (MEMS) accelerometers, gyros and

magnetometers. A configuration of JAGER is shown in

Figure 3. For navigation, JAGER has several different units

that it can carry. There are separate navigation units for

GPS, GLONASS and Locata. For radio frequency (RF)

direction finding, JAGER carries a directional antenna that

is connected to a RF log detector that determines the

received signal strength in the targeted frequency. Rotation

of the JAGER platform results in determination of an

antenna gain pattern. Navigation and jamming signal

reception information is passed to an onboard small form

factor computer – an Intel i7-based Next Unit of

Computing (NUC) to perform jammer direction finding

and localization algorithm.

Figure 3: Fully equipped UAV used during flight tests

Several navigation systems are carried onboard JAGER.

The GPS receiver is based on an U-blox chipset. The

GLONASS receiver utilized a similar chipset. As a result,

it was reasonably easy to integrate with the Pixhawk flight

controller. The GLONASS receiver was the primary source

of navigation during jamming. The system (Figure 4) is

comprised of a GLONASS antenna connected to an U-blox

M8 receiver through a 40 deciBel (dB) L1 band reject filter.

The U-blox M8 receiver was configured to only take and

use GLONASS measurements to create a position solution.

Hence no GPS measurements are used as these could

adversely affect the solution. The 40 dB band reject filter

helped limit the effect of L1 jamming on elements, such as

the automatic gain control (AGC) and low noise amplifier

(LNA), used to support both GPS and GLONASS as these

could affect the GLONASS position solution performance.

Figure 4: GLONASS receiver

A Locata receiver (Figure 5) was also carried for

navigation. The benefit of the system is that it operates on

2.4 GHz, significantly far from GPS interferers, and it

could provide millimeter level truth [4]. We tested

integration both at the GPS laboratory and at the test site.

This is important as RFI and other integration issues pose

a big challenge for a multirotor platform. From testing with

live on air signals, we found that RFI from the command

and control data links affected Locata reception. Not

surprisingly, this was most strong on the ground while next

to the command radio; it dissipated into the air as the

vehicle was farther away from the ground radio. The form

factor of the current Locata receiver also posed an

integration challenge. The metallic housing seemed to

affect the onboard magnetometers. Also, it weighs

approximately 2 kilograms (kg) which reduced our flight

time by half. The testing with Locata diagnosed and

resolved most of these integration issues, however,

opening the way to fully operating JAGER with Locata.

This is potentially important for tests involving more

complex jammers that make a multi-frequency / multi-

constellation solution difficult.

(a) GLONASS antenna (b) 40 dB L1 notch filter

(c) U-blox M8 receiver

Figure 5: Locata receiver mounted to underside of JAGER

To localize the source of the GPS jammer, the sensor being

used is a directional TECOM L1 antenna with 12 dB of

gain and a beamwidth of 35 degrees, depicted more closely

in Figure 6 [5]. In the same figure, the rest of the

components for the sensing can be seen: a bandpass filter

to reduce outside noise and a Linear Technology RF

detector. The RF detector converts signal strength to

voltage, allowing the system to directly measure incoming

signal strength [6].

Figure 6: GPS localization signal strength sensor

JAGER carries many radios for navigation,

communications and operations. Radio navigation sources

are passive and operate on various frequencies: Locata on

2.4 GHz and GPS/GLONASS around 1.6 GHz.

Communication sources are generally two way and operate

at several frequencies including 5.8, 2.4 and 0.9 GHz.

Finally, several sources of RFI exist including the power

system, rotor motors which can draw up to 2 kW of power

and the NUC. The radios are indicated on Table 1.

To mitigate the many sources of RFI onboard JAGER,

copious amounts of copper tape were used for shielding.

This is shown in Figure 7. The shielding helped minimize

the interference from the power sources of the vehicle with

the navigation antennas, minimized interference into the

sensor and reduced interference from the Intel NUC

onboard. Most of the copper shielding was placed on the

platforms between our navigation systems on the top and

our communications on the bottom. This reduced

interference between our 2.4 GHz and Locata and likely

also helped attenuate jamming from the ground to our GPS

and GLONASS receivers.

Table 1: Radio frequencies and RFI sources onboard JAGER

Source Transmit Frequency

Command/Control 2.4 GHz

Telemetry 900 MHz

Datalink 5.8 GHz

Electric Motors RFI

Intel NUC RFI

Figure 7: Copper shielding onboard JAGER to minimize RFI

Safety

Due to the nighttime testing that was performed, in addition

to the normal safety features of the autopilot, the UAV was

augmented with many light emitting diodes (LEDs)

(Figure 8) to assist the pilot in maintaining visual

observation of the UAV in case manual control needed to

be taken at any time during the flight.

Figure 8: LEDs on vehicle for pilot visibility

(a) L1 antenna with 12 dB gain

(b) L1 bandpass filter (c) RF detector

JAMMER PERFORMANCE

The source of jamming for these tests was a COTS GPS

jammer provided by DHS and operated by the test support

personnel at the testing site.

The advertised specifications for the jammer was 0.5 W in

the L1 band with a range of 5-15 m. The jammer range

experienced during flight testing was significantly larger

and differed depending on setting. The jammer

performance over two nights of testing is shown in Figure

9 and Figure 10. In these figures the red lines are where we

did not have a solution output from the U-blox GPS

receiver, and green lines are where we did have a position

solution from the U-blox GPS receiver.

Figure 9: Thursday night GPS jammer performance

In each of these figures, the jammer’s effectiveness is

depicted both on a map and as a percentage of the

measurements with and without a GPS position solution for

specific distances from the jammer. As can be seen, the

effective range of the jammer (the maximum distance

where the GPS receiver was unable to return a position

solution at least once) was approximately 300 m for

Thursday night (Figure 9) and at least 350 m for Friday

night (Figure 10).

Figure 10: Friday night GPS jammer performance

The results suggest a difference in jammer power during

the different nights. One difference was that the jammer

was operated on battery power on Thursday and plugged

into a power source on Friday. Our measurements suggest

that battery powered operation provided less jamming

power than plugged-in operation.

SENSING PERFORMANCE

Previously, JAGER has demonstrated an ability to localize

a Wi-Fi router [1], but for these tests the sensor had to be

changed to be sensitive to the L1 band. With a new sensor

came a new set of challenges, especially with integration

onboard the system and RFI mitigation, which were

discovered during the first nights of testing. While the

sensor measures received signal strength at L1, we

leveraged the rotational motion of a multirotor platform to

generate a sequence of signal strength measurements at

various azimuth angles. This collection of signal strength

measurements was then used to recreate the antenna’s gain

pattern to the jamming source. From there, the bearing to

(a) Percentage of measurements with (green) and without (red)

a GPS position solution as a function of distance

(b) Map of GPS performance

800m

(a) Percentage of measurements with (green) and without (red)

a GPS position solution as a function of distance

(b) Map of GPS performance

300m

the jammer was determined and used in the navigation

systems [1].

The use of the collection of signal strength measurements

to obtain a bearing provides some initial robustness to the

effects of unwanted noise in the sensor, which is described

in this section.

Noise

The signal sensor measured periodic noise spikes. This

resulted in gain patterns such as the one depicted in Figure

11. Despite the noise, an accurate determination of bearing

is still achievable through processing. The cross correlation

(cc) method was particularly good at resolving an accurate

bearing despite the noise. It did make bearing estimation

more challenging for some of the methods employed.

Figure 11: Raw measured gain pattern with noise

These patterns could be filtered out in real time through

thresholding and a nearest neighbor analysis resulting in

the smoother pattern in Figure 12. Note that once the noise

is removed, each of the bearing calculation methods

previously used performs as expected [1].

Figure 12: Filtered gain pattern with sensor noise removed

Compared to our previous testing with Wi-Fi in [1], the

employed sensor and antenna setup has a smaller

beamwidth. The tighter beamwidth improved the accuracy

of the calculated bearing.

Range

Using a very high gain antenna also meant that the system

was able to detect the GPS jammer from great distance.

The sensor clearly detected the jammer at a range of 150 m

from the jammer (Figure 13) with a pattern that was good

enough to have bearing extracted from any one of the

bearing calculation methods used.

Figure 13: Filtered gain pattern at 150 m from jammer

We were able to test the sensor to a distance of 800 m from

the jammer, resulting in the pattern shown in Figure 14.

Note, in this pattern much more noise is present and the

pattern no longer clearly resembles the expected gain

pattern. Therefore several of the methods are unable to

determine bearing correctly, however the more basic

method (using maximum value) is still able to perform well

and determine the bearing. The cross correlation method

provides a correlation coefficient that indicates how well

of a correlation was performed. For the measurements far

from the jammer, the cc coefficient is very small and allows

the system to use one of the more basic methods [1].

Figure 14: Filtered gain pattern at 800 m from jammer

Due to battery limitations on JAGER for these tests we

were not able to perform range tests beyond 800 m,

however the RF detector is sensitive down to -65 dB

relative to a milliWatt (dBm), leaving approximately 7

dBm of margin left at 800 m. Therefore we are confident

this system would be able to detect the jammer from

beyond 800 m; however, the ability to extract accurate

bearing would be more limited.

LOCALIZATION

For localizing the GPS jammer, our goal was to not only

test our algorithms with a GPS jammer, but also to test at

longer ranges than we have previously tested. For these

tests, the initial start point was 350 m away from the

jammer and the search area used was a 500 m x 500 m

search area more representative of a possible scenario than

the 100 m x 100 m search areas previously used [1]. We

represent the square search area as an n by n set of discrete

cells. JAGER maintains a belief or probability distribution

over these cells. The weight of each cell is the probability

the jammer is in that cell. Every time JAGER rotates and

makes a bearing measurement, this belief is updated using

Bayes’ rule and a measurement model. This type of filter is

called a discrete Bayes’ filter or sometimes a histogram

filter [7].

The measurement model for our system is the true bearing

from JAGER to the jammer with additive Gaussian noise.

This noise and errors result in the bearing accuracy having

a standard deviation of roughly 10 degrees [1].

The planning problem consists of selecting a new

rotation/measurement location given the belief. In our

previous work, we explored multi-step planning with

POMDPs [1]. Although POMDPs offer a principled

approach to decision making under uncertainty [8], they

are computationally intractable in the general case [9]. In

previous work, we used small search areas (100 m x 100

m) that were coarsely discretized (11 x 11 grids) [1], so we

could apply approximation techniques yielding good

solutions.

In these experiments, we used more realistic search areas.

This requires grids larger than 11 x 11 so individual grid

cells are not too large, giving us the precision we desire.

Therefore, we explore one-step optimal also known as

greedy solutions that minimize some information-theoretic

objectives over a planning horizon of one. The following

subsections describe different greedy planners used.

Entropy Minimization

A common method for localization is minimizing the

expected entropy after the next measurement [10]. Entropy

is a measure of spread in a distribution. A uniform

distribution maximizes entropy and one in which all weight

is concentrated in a single cell has zero entropy.

One way to compute expected entropy is to discretize the

observation space into a discrete number of observations.

Given JAGER’s position and the jammer’s position, we

can assign probabilities to each observation. Each

observation yields a new belief whose entropy we can

compute. Because we do not know which observation we

will receive, we take an expectation over all possible

observations. We also do not know the jammer’s location,

so we take an expectation over our belief. This provides the

entropy given a position for JAGER. To find the best new

position, this process is repeated for every position to

which JAGER can travel. The set of possible JAGER

positions contains the center of every cell in our search

area. That is, JAGER moves to a new position xv,t+1

according to the following equation:

𝑥𝑣,𝑡+1 = argmin𝑥𝑣

𝐸𝑥𝑗 ∈ 𝑏𝑡

𝐸𝑦𝑡+1𝐻(𝑏𝑡+1)

where bt is the belief at time t, xj denotes a possible jammer

location, and H(bt) is the entropy of belief bt. This

computation requires iterating over all grid cells twice,

leading to computational complexity of O(|Y|n4), where |Y|

is the number of discrete observations. These computations

become intractable when the number of grid cells per side,

n, increases.

We computed controls online for 25 x 25 grids on a 500 x

500 m search area, yielding reasonable 20 x 20 m grid cells.

Any discretization more fine was difficult to compute in

less than one second during flight. Figure 15 shows an

example of a localization run carried out using this method.

After each measurement, the belief concentrates around the

jammer location (red triangle).

Figure 15: Entropy minimization belief distribution at each step (a) through (g) and overall flight path of search (h)

Determinant Minimization

As mentioned in the last subsection, computing entropy

minimization can become intractable as the number of grid

cells in a discrete filter increases. If we want very fine cells,

for example half a meter per side, we need to make some

approximations.

A common approximation in estimation is linearizing the

measurement model and assuming beliefs are roughly

Gaussian. Filtering is computationally inexpensive with

these approximations. More importantly, estimating the

covariance after future measurements becomes far more

tractable. Kalman filters struggle in bearing-only

localization schemes because the measurement function is

very nonlinear, however. Our solution is to maintain a

discrete filter for localization, but fit a Gaussian

approximation after each step. This approximation can be

calculated from the mean position and covariance of cells

in our discrete belief.

An interesting feature of Kalman filters is that future

covariance can be estimated easily given future

measurement locations and the estimate mean. This feature

becomes most apparent when using the information filter –

a Kalman filter that uses the canonical form for Gaussians:

𝜉 = 𝛴−1𝜇

𝛺 = 𝛴−1

When using the canonical form, the update step in the

Kalman filter becomes:

𝛺𝑡+1 = 𝛺𝑡 +𝐶𝑡𝑇𝐶𝑡𝜎2

where Ct are the linearized measurement dynamics for the

target mean at time t and σ is the bearing error standard

deviation (10 degrees in our case). This gives us the inverse

of our covariance matrix at future step, without discretizing

the observation space and iterating over it. This is not a true

expectation over jammer locations, because the mean

estimate is used, but the computational trade-off is

favorable. Computing the covariance for a measurement

from a future measurement location is now O(n2), a great

reduction from the entropy reduction.

Once the covariance of a measurement from a new position

is estimated, it is possible to iterate over all possible new

measurement locations. To convert the covariance to a

scalar metric, the determinant can be used. Minimizing the

determinant is equivalent to minimizing the area of an

uncertainty ellipse representing the belief.

Figure 16 shows this method in a flight test. The mean and

95 percent confidence ellipse after each step is shown.

JAGER selects new positions that are almost identical to

the mean location. This might be because the linearized

bearing measurement equation approaches infinity as the

relative distance between the jammer and JAGER

decreases. It is also a possible use of the determinant as a

(a) (b) (c) (d)

(e) (f) (g) (h)

metric causes JAGER to move to the mean estimate. In

bearing localization, determinant minimization sometimes

moves sensors toward the estimate rather than

perpendicularly, which is intuitive. This is because the

determinant corresponds to the area of an uncertainty

ellipse. This area can be small if uncertainty is large in one

direction but very small in another.

Figure 16: Determinant minimization belief state at each step (a) through (f) and overall flight path of search (g)

Maximum Eigenvalue Minimization

In some cases, it is desirable to move towards the jammer.

For example, if we wanted to photograph the jammer,

moving to the belief mean would make sense. However, we

often want to maintain some distance from the jammer, as

measurements near it can be noisier [1]. Therefore, we

explore another metric using the approximation approach

mentioned in the previous subsection.

Instead of minimizing the determinant, we can minimize

the largest eigenvalue of the future covariance matrix.

Instead of minimizing the area-like determinant, this

operation minimizes the largest axis of an uncertainty

ellipse representing our belief. This is often done to

minimize the worst-case uncertainty along any axis.

The eigenvalue minimization approach was flown and an

example trial can be seen in Figure 17. JAGER often elects

to take new measurements from a position perpendicular to

the largest axis of the uncertainty ellipse. This behavior

makes sense in bearing localization. Interestingly, JAGER

also tended to stay some distance away from the jammer.

This can be desirable as perhaps GPS will be available for

navigation farther from the jammer.

(a) (b) (c) (d)

(e) (f) (g)

Figure 17: Eigenvalue minimization belief distribution at each step (a) through (f) and overall flight path of search (g)

NAVIGATION PERFORMANCE

Three different navigation systems were onboard JAGER:

GLONASS, GPS and Locata. GLONASS was the primary

navigation system used throughout all flight tests and

performed very reliably. GPS was onboard to analyze both

the performance of the jammer and the possibility of

navigating on GPS even when attempting to localize a GPS

jammer. JAGER was also equipped with Locata, a ground

based positioning system, to allow for on-ground and in-air

integration testing for potential future use on JAGER.

GLONASS

GLONASS was the main method used to navigate the

UAV during the L1 GPS jamming exercise. Because the

jammer was limited to the L1 band, and because the

frequency of GLONASS was just outside the range of the

jammer, GLONASS proved to be a reliable navigation

system that was small and integrated easily with our flight

controller. Figure 18 shows the horizontal (EPH) and

vertical (EPV) position covariance for testing on Friday

night. It can be seen that the horizontal covariance was

almost always under 4 m, which was typical each of the

nights. The only degraded performance experience with

GLONASS was when the system was turned on. This

occurred in close proximity to the jammer and is therefore

most likely caused by the jammer.

Figure 18: GLONASS horizontal and vertical position solution

covariance in meters Friday night

GPS

The availability of a solution from the GPS receiver was

very high despite the jammer being active (Figure 9).

Further inspection showed that while a receiver solution

was available, it was not necessarily of high quality and

could have significant error. Figure 19 shows the GPS and

GLONASS solution in blue and green, respectively. Here

it can be seen that even though the system was getting a

valid GPS position, the position difference from

GLONASS were very large.

(a) (b) (c) (d)

(e) (f) (g)

Figure 19: GPS position error in flight with GPS L1 jammer

active

A plot of the GPS covariance can be seen in Figure 20. This

figure shows that the covariance of the position solution for

GPS was much closer to the 10 m mark than the 2-3 m seen

with GLONASS. For best operation of the autopilot, the

errors need to be 5 m or better.

Figure 20: GPS horizontal and vertical position solution

covariance in meters Friday night

Looking more closely at the covariance of the GPS solution

and moving from simply a binary check of whether or not

a position solution was calculated to a tiered analysis

results in Figure 21 and Figure 22. In these figures, red

again means no position solution was received, purple

represents a GPS solution with horizontal covariance of

more than 10 m, yellow a covariance in the 5-10 m range

and finally green representing a GPS position solution with

a covariance of less than 5 m.

Figure 21: Tiered GPS performance Thursday night

Dividing the GPS position results into these four different

categories shows that the effective range of the jammer is

about 100 m more than initially observed. The results show

that one needs to use GPS with care near jamming,

especially with non-certified GPS receivers. For such

receivers, jamming can potentially pose a more severe

threat than just loss of GPS position. It could present

misleading information. This is a similar experience seen

by the General Lighthouse Authorities in their GPS

jamming trials [11].

Because our sensor is able to “see” the jammer from such

a great distance, this will help in the navigation challenge

allowing us to potentially rely on GPS for part of the flight

and changes the requirements of the denied navigation

system.

25

0m

(a) Percentage of measurements in each performance category

as a function of distance

(b) Map of GPS performance

800m

Figure 22: Tiered GPS performance Friday night

Locata

We were able to successfully integrate Locata into the open

source Pixhawk autopilot system to obtain solutions both

while on the ground and in the air. JAGER was able to fly

the vehicle autonomously with Locata. The weight of the

payload combined with the rest of the equipment reduced

flight test time by half, however. Hence, it was not used as

the primary positioning system used. It provides a very

accurate truth measurement, however, and its frequency

separation from any of the GNSS signals makes it a

compelling positioning solution for future testing of a

wider range of jammers.

CONCLUSION

The JAGER tests at DHS FREJE were very successful. We

accomplished our major test points. Under the scenario of

a single static jammer in an open field, JAGER

successfully located the jammer autonomously from

various distances and using several different algorithms.

JAGER reliably determined the direction of the jammer

from at least 800 m away, which is well beyond the

effective range of the jammer tested. Despite the shorter

horizon of these localization algorithms, they all very

rapidly localized the jammer and consistently localized the

jammer to within an appropriate distance. While we grew

the search area by 25 times from previous tests, the

localization algorithms only took on average

approximately 1.5 times the number of steps to

successfully localize the jammer, which is very promising

for JAGER’s ability to rapidly localize a jammer in a large

search area.

GLONASS and Locata were both used for navigation. Our

navigation test showed that with the jammer and the

GLONASS setup used, JAGER was able to have

GLONASS navigation with high availability in the air. In

close proximity to the jammer, the receiver experienced

difficulties quickly acquiring GLONASS. This is likely

due to the jamming affecting elements used by both GPS

and GLONASS, such AGC and LNA. The 40 dB band

rejection is not enough when right next to the receiver

leading to the jamming power affecting the LNA and/or

AGC. Locata functioned both in the air and on the ground.

JAGER flew autonomously using Locata. Examination of

our data showed that GPS L1 solution was often available

in the air. JAGER airframe likely provided some mitigation

as a layer of copper tape was used to shield the top shelf

with GPS and other navigation antennas from the lower

two shelves as well as the ground below.

The results demonstrated that JAGER can locate a jammer

from a distance that is greater than effective jamming

radius on JAGER and perhaps other GNSS equipment.

This result is profound in that it means that JAGER could

navigate on GNSS from a standoff distance (where GNSS

is not affected) and still localize the jammer given an

appropriate algorithm.

FUTURE WORK

JAGER operated under very favorable scenarios at DHS

FREJE. It operated in an open environment with a single

static jammer. We limited the complexity of the testing as

this was the first test of JAGER with actual rather than

simulated jamming. JAGER will need to operate under

much more complicated scenarios where there may be

multiple jammers and/or moving jammers. Additionally,

JAGER is meant to operate with suburban and urban

environments such as near an airport. These environments

have buildings and other structures that can cause signal

blockages and multipath. This makes localization more

challenging. We are building upon our existing research,

hardware and software to develop the capability to handle

these more challenging scenarios.

(a) Percentage of measurements in each performance category

as a function of distance

(b) Map of GPS performance

300m

Navigation in GNSS denied environments is an important

component on JAGER. Several possibilities exist. As

demonstrated from our tests, GNSS may have some

utility. There are regions where both GNSS is available and

jamming is detectable. This may be coupled with a vision

system to provide relative navigation should JAGER need

to enter a more severely jammed region. This is one area of

future work as we were not able to test the vision

system. Despite being able to use GNSS, JAGER should

have navigation means that is robust to GNSS RFI.

ACKNOWLEDGEMENTS

The authors gratefully acknowledge the support of the

Federal Aviation Administration (FAA) and the Stanford

Center for Position, Navigation and Time (SCPNT). We

thank DHS for supporting these trials. We also want to

thank Erik Lunderg of MITRE for his assistance on using

GLONASS during jamming. We also thank Locata for

their support in integrating the use of the Locata

positioning system onboard JAGER. Finally, we thank the

support personnel during the test trials who were integral

in our successful testing.

DISCLAIMER

The views expressed herein are those of the authors and are

not to be construed as official or any other person or

organization.

BIBLIOGRAPHY

[1] A. Perkins, L. Dressel, S. Lo and P. Enge, "Antenna

Characterization for UAV Based GPS Jammer

Localization," in Proceedings of the 28th

International Technical Meeting of The Satellite

Division of the Institute of Navigation (ION GNSS+

2015), Tampa, 2015.

[2] John Hopkins Applied Physics Laboratory, "GPS

Risk Assessment Study - FInal Report," Jan 1999.

[3] Pixhawk, "PX4 autopilot project," [Online].

Available: https://pixhawk.org/.

[4] A. Trunzo, R. Ramirez and J. Baldwin, "The

UHARS Non-GPS-Based Positioning System

(NGBPS)," in Proceedings of the 27th International

Technical Meeting of The Satellite Division of the

Institute of Navigation (ION GNSS+ 2014), Tampa,

Florida, 2014.

[5] TECOM, "Circularly Polarized Directional Antenna

Datasheet," [Online]. Available: http://www.tecom-

ind.com/files/1/536146a0d9e3c-

WebDA0206010.pdf.

[6] Linear Technology, "RF Power Detector

Datasheet," [Online]. Available:

http://cds.linear.com/docs/en/datasheet/5538f.pdf.

[7] S. Thrun, W. Burgard and D. Fox, Probabilistic

robotics, MIT Press, 2005.

[8] L. P. Kaelbling, M. L. Littman and A. R. Cassandra,

"Planning and acting in partially observable

stochastic domains," Artificial Intelligence, vol.

101, no. 1-2, pp. 99-134, 1998.

[9] O. Madani, S. Hanks and A. Condon, "On the

undecidability of probabilistic planning and infinite-

horizon partially observable Markov decision

problems," in AAAI, 1999.

[10] G. M. Hoffmann, S. L. Waslander and C. J. Tomlin,

"Distributed cooperative search using information-

theoretic costs for particle filters, with quadrotor

applications," in Proceedings of the AIAA

Guidance, Navigation, and Control Conference and

Exhibit, 2006.

[11] A. Grant, P. Williams, N. Ward and S. Basker,

"GPS Jamming and the Impact on Maritime

Navigation," Journal of Navigation, vol. 62, no. 2,

pp. 173-187, April 2009.